RTX 3070 compatibility with Pytorch
Asked Answered
B

1

5

NVIDIA GeForce RTX 3070 with CUDA capability sm_86 is not compatible with the current PyTorch installation. The current PyTorch install supports CUDA capabilities sm_37 sm_50 sm_60 sm_70.

So I'm currently trying to train a neural network but I'm getting this issue. It seems that the GPU model I have is not compatible with the version of PyTorch that I have.

The output of my nvcc -V is:

Cuda compilation tools, release 11.4, V11.4.48
Build cuda_11.4.r11.4/compiler.30033411_0

and my PyTorch version is 1.9.0.

I've tried changing the CUDA version from what was initially 10 to 11.4 and there was no change. Any help would be hugely appreciated.

Butterfield answered 26/7, 2021 at 11:44 Comment(4)
The problem has nothing to do with your CUDA version (although I doubt you have the correct version for the Pytorch build you are using). The probably is that the PyTorch developers are not releasing builds which support your GPU because it is too new. Your choices are either to wait until they do support your GPU model. or build your own PyTorch from source with support for your GPU model included, if that is possibleImpresario
PyTorch compatibility matrix suggests that pyTorch 1.9 can be configured for CUDA 11.1, which may allow you to run with RTX 3070. It says to run conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c conda-forge and has a note conda-forge channel is required for cudatoolkit 11.1.Anaya
This is likely a result of installing pytorch for the wrong cuda version. To use a compute capability 8.6 GPU you must install the 11.1 version of pytorch since compute capability 8.6 is only compatible with cuda >= 11.1. Once you install the proper build, if you are still having problems please post the result of print(torch.cuda.version) and print(torch.cuda.get_arch_list())Adelina
github.com/pytorch/pytorch/issues/45028Holcombe
P
8

It might be because you have installed a torch package with cuda==10.* (e.g. torch==1.9.0+cu102) . I'd suggest trying:

pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html
Ppi answered 28/1, 2022 at 6:40 Comment(0)

© 2022 - 2024 — McMap. All rights reserved.